FanGraphs Now Featuring coolstandings.com Playoff Odds
I’m pleased to announce that earlier this year, coolstandings.com joined the FanGraphs family and we’ve now integrated their playoff odds into FanGraphs, with some additional features available here on our site as well.
I’ve always been an admirer of the work Greg Agami and Sean Walsh did at coolstandings.com, and we couldn’t be more pleased to have their product as a part of FanGraphs. And we’re excited about using some of the data available on FanGraphs to make the forecasts even better.
There are currently three different ways we simulate the season to generate the playoff odds:
FanGraphs Projections Mode – This mode uses the combination of Steamer and ZiPS projections, along with the updated FanGraphs Depth Charts, to calculate the expected winning percentage of each remaining game in the major league season. This model incorporates the latest updates to each team’s roster, including injuries, acquisitions, and adjusted playing time allocations.
Season to Date Stats Mode – This mode uses current season to date statistics, weighted more heavily towards the most recent games, to calculate the expected winning percentage of each remaining game in the major league season. (This is the old “Smart” mode on coolstandings.com)
Coin Flip Mode – This mode uses a coin flip to determine the winner of each game, giving each matchup 50/50 odds. (This is the old “Dumb” mode on coolstandings.com)
The coolstandings models incorporate actual remaining schedules into their forecasts, so in addition to the estimates of each team’s talent, the strength of their upcoming opponents is also included, and the simulations are run 10,000 times hourly.
In addition to the standard odds of winning the division or earning a place in the wild card play-in game, we also have the odds of each team winning each level of the playoffs all the way up to odds of winning the World Series.
Right now we’re showing the current playoff odds, but you can visit coolstandings.com for all the historical playoff odds. We’ll also be adding these to FanGraphs in short time. You can find a permanent link to the Playoff Odds from the Standings drop down in the toolbar at the top of the site.
David Appelman is the creator of FanGraphs.
Great! Any reason why coolstandings is used and not baseballprospectus? Do they calculate the odds in the same way?
FanGraphs Projection mode seems to be, by far, the most well-informed of these three systems. I wonder if you could take a moment to explain what the advantages of the coolstandings methods are over the FG one.
The FanGraphs projections should be the most predictive, but we wanted to bring over both the “Smart” and “Dumb” systems that were on coolstandings.com as well. The coolstandings.com original systems can also be run historically and we plan to bring over all the historical data that is available on coolstandings.com.
Chinese people shouldn’t be allowed to play baseball.
I appreciate coolstanding’s diligence in running all those simulations, but recent performance is way overweighted in their calculations. One win or loss in the last game makes a big difference.
Can you make them change this?
On FanGraphs we are now using our projections to power the coolstandings.com projections in addition the other other methods. At this point in the season, 1 win or loss is going to make something of a difference in terms of playoff odds anyway.
…and isn’t the second one just weighting small sample sizes over larger ones? What is the point of that? (Maybe something about new players joining the team, or players making mechanical changes?)
Just a comment. Hovering over the NLDS and NLCS columns brings up a tooltip mentioning the ALDS and ALCS respectively.
That should be fixed now!
Nice addition to the site, guys. Does the DS/CS/WS odds calculator take into account who the starting pitchers would potentially be and simulate for each game that could be in each round of the playoffs? Also does it weight probabilities for each potential team v team playoff matchup? Cheers!
For the playoffs it does not take into account the playoff rotations. It’s something that would be neat to do, but it gets complicated trying to simulate playoff rotations.
We sim each season in its entirety, so more common matchups will have more weight in the post-season odds, but there is no other weighting applied. We could spit out which matchups are most likely to happen, but we’re not saving that data right now.
Now that you have these playoff odds, would you be able to tie them to WPA, such that there would be a PPA stat? I know this wouldn’t really be a meaningful stat, but it would be incredibly cool.
Yes, this is possible to do, though I’m not sure what the timeline would be. Dave Studeman did some great research on this at the 2012 Phoenix SABR Analytics Conference that went through history and did PPA for all players in the post-season. Really fun stuff.
Nice, I’ll have to look into that. Thanks for adding the playoff odds…just realized I didn’t say that in the first post.
Dave Studeman did this in 2009 at THT if you’re interested.
http://www.hardballtimes.com/main/article/postseason-probability-added/
Awesome. This is the only thing the Projected Standings were missing.
One comment. I see that under the Projected Standings that teams ROS Win% do not match the ROS Wins and Losses. The Wins & Losses line up with the CoolStandings Playoff odds, but Win% does not. For example, the Red Sox are expected to be 17 and 12 the rest of the way (.586), but the ROS Win% says .601. Why the variance. Thanks
The difference is that the projected standings page does not take into account strength of schedule, where the playoff odds do.
Why not include strength of schedule in the projected standings page? Or do you have plans to do this soon?
Right now there are no plans to incorporate strength of schedule into the projected standings. So, if you want strength of schedule use the playoff odds, if you don’t, use the projected standings. There are definitely uses for both.
The rest-of-season Win Percentage is different from the Depth Charts – e.g. Nationals are .582 on the Playoff Odds page and .539 on the Depth Charts. Is the difference strength of schedule?
Yep, strength of schedule is the difference!
So by the measure the Cardinals have the hardest schedule down the stretch in MLB, and the Pirates have the second easiest. Very cool
No subtracted wrong columns. Braves easiest schedule, O’s hardest schedule.
Is it possible with these simulations to give confidence intervals (or some other distributional information beyond playoff odds) for the RoS projections?
Just last week I made a spreadsheet to calculate the Indians’ playoff odds using Fangraphs projections! This is much better, thank you.
One thing I had fun looking at was playoff odds based on various RoS Win%s. Gives you a good sense of how well a team needs to play to give themselves a good chance of making the playoffs. Any plans to do something like that?
This is awesome. Now we can track daily changes in the WS% to figure out the WCPA (World Championship Probability Added) for each game.
Playoff odds stats have never sat right with me. It’s something to do with how early teams who are at the top of the standings are said to have a 98% or above chance of making the playoffs. For instance, the Braves right now are said to have a 99.9% chance of winning the division, and a 100% chance of making the playoffs, and yet I can envision a scenario (unlikely, yes, but not impossible) in which they do neither. The odds don’t seem designed to account for in-season personnel changes, whether due to injury or trades that work out particularly well or poorly. And the odds tend to get talked about primarily in retrospect (they blew a 98.8% chance of making the playoffs!…well, yes, their chances did bubble up to 98.8% that one day in early August), not as a tool of utility in the moment.
I believe it is just rounding to 100%. Of course until you actually clinch your chances of making the playoffs are never 100.000000000000%
The probabilities are perhaps a little extreme (i.e. too close to 100%) but not ridiculously so. But I do agree that early in the season (say April through June), the alleged playoff probabilities calculated by most of these websites (BPro in particular; I haven’t paid as much attention to CoolStandings) had been way too extreme. Basically they failed to do enough regression to the mean. However BPro re-tooled their formulas this season and seemed to have less extreme early-season probabilities.
This is great. I thought it was silly earlier this year when the Braves had 50 games left and 43 were against teams with losing records, but were projected to go 27-23.
Does the strength of schedule calculation include off days? I know the Red Sox are facing some good teams down the stretch, but also have a number of days off, so just going on strength of opponent might make it look worse than it really is for them. Also, thanks! This is another great tool to add to the site.
How is it predicting the win expectancy of each game given the input projections. Is there a previous post somewhere that explains this methodology?
For example, are you morphing all five starting pitchers and all bullpen members together to create one same pitching input for every single game a team plays?
Is there a HFA adjustment? If so, what is it?
Is defense taken into account?
why is WS odds not half AL/NLCS odds in coin-flip mode?
Do the simulations include the possibility of multiple significant injuries happening to one team at one time?
According to Fangraphs Mode, the AL team seems to be about a 2:1 favorite in the World Series. Am I reading that right?
Would the projected odds be different if team talent wasn’t a point estimate, but rather a range?
For example, Boston has the talent of a .601 team (according to the standings), leading to an expected win percentage of .581 for the rest of the season (discrepancy owing to SOS, apparently). But what if you assumed a range of talent levels for Boston between .561 and .641, say? Probability-weighted, presumably, and likely regressed, given the greater likelihood of .561 over .641. Would their odds still be 98.1%, or lower? I would have thought lower, but I’m not sure.
Can you allow the user to choose Steamer alone as the projection source? As has been shown by others, ZIPS rest-of-season projections are simply not very good.
This is probably a dumb question – math was never my strong suit – but do any of the projection methods indicate clearly when a team has been “mathematically eliminated” from the playoffs? I see a bunch of teams with 0.0% chances, but does that mean it is mathematically impossible to reach the playoffs, or just that the projections think they have no chance, even if it’s not mathematically impossible?